Abstract
Image recognition systems in automotive applications should be robust. The systems should maintain the recognition ratio even if a part of the image captured from in-vehicle cameras is hidden by trees or other obstacles. A new robust recognition algorithm can be implemented on an inexpensive chip is proposed. To make the circuit simple and to increase the operational speed, it adopts a template matching using Levenshtein distance. The algorithm consists of only integer-operations and can be mounted on an inexpensive FPGA chip. A hardware module for the algorithm has 66,560 cells and is implemented on Xilinx Spartan-3 chip using Bach-Cat 49.8 MHz. The algorithm achieves an execution time of 17.0 msec at low power. Recognition results show 20% improvement compared with the conventional pattern matching algorithm.
KEYWORDS – Recognition, Road Sign, Parallel processing, FPGA, Levenshtein distance